Summary

• Problem: NVIDIA onboarding/model-selection surfaces used only a bundled static catalog, so newly promoted build.nvidia.com models needed an OpenClaw release before appearing.
• Why it matters: new NVIDIA users should see NVIDIA's current ranked featured models during setup instead of stale defaults.
• What changed: the NVIDIA provider now fetches NVIDIA's public featured-model catalog, validates it, caches successful results for 24 hours, promotes ranked rows ahead of bundled fallback rows, and keeps the manifest catalog as the offline/static fallback.
• What did NOT change (scope boundary): no core provider discovery behavior changed, no credentials are sent to the public catalog endpoint, and existing NVIDIA provider config/onboarding fallback defaults remain available.

Change Type (select all)

• Bug fix
• Feature
• Refactor required for the fix
• Docs
• Security hardening
• Chore/infra

Scope (select all touched areas)

• Gateway / orchestration
• Skills / tool execution
• Auth / tokens
• Memory / storage
• Integrations
• API / contracts
• UI / DX
• CI/CD / infra

Linked Issue/PR

• Closes [Feature]: NVIDIA Featured Models API integration #79431
• Related #
• This PR fixes a bug or regression

Real behavior proof (required for external PRs)

• Behavior or issue addressed: OpenClaw can consume NVIDIA's public featured-model feed and promote ranked catalog rows while retaining a bundled fallback.
• Real environment tested: local OpenClaw checkout on Linux with Node 22 from .local/node22/bin.
• Exact steps or command run after this patch: PATH=/home/ubuntu/.local/node22/bin:$PATH node -e 'fetch("https://assets.ngc.nvidia.com/products/api-catalog/featured-models.json").then(async r=>{const j=await r.json(); console.log(JSON.stringify({status:r.status, count:j["featured-models"]?.length, first:j["featured-models"]?.[0]}, null, 2));})'
• Evidence after fix: terminal output from the live NVIDIA public catalog fetch:

$ PATH=/home/ubuntu/.local/node22/bin:$PATH node -e 'fetch("https://assets.ngc.nvidia.com/products/api-catalog/featured-models.json").then(async r=>{const j=await r.json(); console.log(JSON.stringify({status:r.status, count:j["featured-models"]?.length, first:j["featured-models"]?.[0]}, null, 2));})'
{
  "status": 200,
  "count": 4,
  "first": {
    "model": "nemotron-3-super-120b-a12b",
    "model-name": "Nemotron 3 Super 120B",
    "context": 262144,
    "max-output": 8192
  }
}

• Observed result after fix: the live endpoint schema matches the provider parser contract, and the patched provider catalog promotes ranked rows ahead of bundled fallback rows.
• What was not tested: a full interactive onboarding run with a real NVIDIA API key; the change is covered at the provider catalog/plugin hook level.
• Before evidence (optional but encouraged): current main documents and implements NVIDIA as a static bundled catalog.

Root Cause (if applicable)

• Root cause: N/A, feature request for live NVIDIA featured-model discovery.
• Missing detection / guardrail: N/A.
• Contributing context (if known): the NVIDIA plugin previously exposed only manifest-backed static model rows.

Regression Test Plan (if applicable)

N/A

• Coverage level that should have caught this:
  • Unit test
  • Seam / integration test
  • End-to-end test
  • Existing coverage already sufficient
• Target test or file: extensions/nvidia/provider-catalog.test.ts, extensions/nvidia/index.test.ts
• Scenario the test should lock in: ranked featured rows are promoted, bare NVIDIA model ids are normalized with the NVIDIA vendor namespace, unavailable featured feeds fall back to bundled rows, and successful featured fetches are cached.
• Why this is the smallest reliable guardrail: the behavior is provider-owned catalog construction plus plugin catalog augmentation, so provider-local tests cover the contract without a full gateway run.
• Existing test that already covers this (if any): N/A.
• If no new test is added, why not: N/A.

User-visible / Behavior Changes

NVIDIA setup and model-selection surfaces can show NVIDIA's current public featured models without an OpenClaw code update. If the public feed is unavailable or malformed, users still see the bundled fallback catalog.

Diagram (if applicable)

Before:
NVIDIA provider catalog -> bundled manifest rows only

After:
NVIDIA provider catalog -> public featured feed -> ranked rows + bundled fallback rows
                        -> fetch/validation failure -> bundled fallback rows

Security Impact (required)

• New permissions/capabilities? (Yes/No) No
• Secrets/tokens handling changed? (Yes/No) No
• New/changed network calls? (Yes/No) Yes
• Command/tool execution surface changed? (Yes/No) No
• Data access scope changed? (Yes/No) No
• If any Yes, explain risk + mitigation: the NVIDIA provider performs a credential-free GET to NVIDIA's public featured-model JSON endpoint. The response is schema-validated, bounded by a short timeout, cached only after successful parsing, and falls back to bundled static rows on failure.

Repro + Verification

Environment

• OS: Linux
• Runtime/container: Node 22 via .local/node22/bin, local checkout
• Model/provider: NVIDIA provider
• Integration/channel (if any): N/A
• Relevant config (redacted): no secrets required for public catalog proof

Steps

1. Fetch NVIDIA's public featured-model JSON endpoint and inspect the returned status/count/first row.
2. Run targeted NVIDIA provider tests.
3. Run staged changed-surface validation.

Expected

• NVIDIA's endpoint returns ranked featured model metadata.
• Provider tests pass for live-row promotion, fallback behavior, and cache reuse.
• Staged changed gate passes for extension/docs lanes.

Actual

• NVIDIA endpoint returned status: 200, count: 4, first row nemotron-3-super-120b-a12b with context 262144 and max output 8192.
• pnpm test extensions/nvidia/provider-catalog.test.ts extensions/nvidia/index.test.ts extensions/nvidia/onboard.test.ts passed: 3 files, 17 tests.
• pnpm check:changed --staged passed for lanes extensions, extensionTests, docs.

Evidence

Attach at least one:

• Failing test/log before + passing after
• Trace/log snippets
• Screenshot/recording
• Perf numbers (if relevant)

Human Verification (required)

What you personally verified (not just CI), and how:

• Verified scenarios: live NVIDIA endpoint shape, ranked featured-row promotion, fallback to bundled rows on HTTP failure, cache reuse, plugin augmentModelCatalog fallback and live rows, extension typecheck/lint/import-cycle changed gate.
• Edge cases checked: malformed/unavailable endpoint path via HTTP 503 fallback; bare NVIDIA model id normalization from nemotron-... to nvidia/nemotron-...; duplicate fallback row suppression.
• What you did not verify: full onboarding against a real NVIDIA API key.

Review Conversations

• I replied to or resolved every bot review conversation I addressed in this PR.
• I left unresolved only the conversations that still need reviewer or maintainer judgment.

If a bot review conversation is addressed by this PR, resolve that conversation yourself. Do not leave bot review conversation cleanup for maintainers.

Compatibility / Migration

• Backward compatible? (Yes/No) Yes
• Config/env changes? (Yes/No) No
• Migration needed? (Yes/No) No
• If yes, exact upgrade steps: N/A

Risks and Mitigations

• Risk: NVIDIA's public catalog endpoint is temporarily unavailable, slow, or changes schema.
  • Mitigation: fetch uses a short timeout, validates the expected fields, caches successful results for 24 hours, and falls back to the bundled manifest catalog on any failure.